review-anvil-readonly
Warn
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a bash command using user-supplied arguments (
$ARGUMENTS) to check parameter pins. This pattern carries a risk of command injection if the arguments are not correctly sanitized or escaped before shell execution. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external source code from repositories or pull requests. This code could contain malicious instructions intended to influence the agent's behavior or manipulate the final review report. Evidence Chain: 1. Ingestion points: The
targetparameter inSKILL.md. 2. Boundary markers: No specific delimiters or 'ignore' instructions are defined for the ingested code in this preset. 3. Capability inventory: File system reads, local artifact creation, and execution of aproof_runner. 4. Sanitization: Mentions an optional 'adversarial' review mode to filter out unsafe suggestions. - [DYNAMIC_EXECUTION]: The skill utilizes a
proof_runnerto execute 'generated probes' during the review. While isolation measures are described, the execution of code synthesized from potentially untrusted inputs remains a security risk.
Audit Metadata